Flexible, Conformal Phased Arrays with Dynamic Array Shape Self- Calibration

Austin C. Fikes 1, Amirreza Safaripour, Florian Bohn, Behrooz Abiri, and Ali Hajimiri Caltech Holistic Integrated Circuits Lab, Caltech, Pasadena, CA 91011, USA [email protected]

Abstract— Flexible and conformal phased arrays enable a mechanical resistive sensors are used to measure shape broad range of novel applications. One of the major challenges for deformation. The sensors add an additional interface, such systems is that they experience a change in their behavior significantly increasing the complexity of the overall system. when bent or deformed. A self-calibrating flexible These drawbacks limit the scalability of the array and prevent can overcome this by estimating the relative position change of its generic arrays from adopting the calibration scheme. elements as they undergo local deformations. In this work, we demonstrate a dynamically flexible and conformal 8-element This work presents a flexible 8-element array with transmit phased array based on a custom CMOS transceiver unit. Beam- and receive capability and proposes a self-contained shape steering is demonstrated with the flexible array flat and with the calibration method using only the coupling between elements array conformed to convex and concave bend radii of ±120 mm. in the array. By using the transmit and receive capability to In addition, we propose and test a shape calibration method that measure the coupling between elements in the array, uses only the coupling between elements, using the flexible phase mechanical deformation of the array is measured. array. Keywords—flexible electronics, phased array, integrated II. FLEXIBLE PHASED ARRAY circuits, integrated circuits, calibration. A. System Overview I. INTRODUCTION The presented 1-D phased array consists of 8 Flexible, lightweight, and conformal phased arrays open a transmitter/receiver custom CMOS integrated circuits plethora of new applications ranging from RF active wearables operating at 10 GHz, each paired with a radiator. All to ultra-light weight space applications and many things in transceivers are phase locked to a shared reference signal. Fig. between. Despite their tremendous potential, they present a 1 shows the flexible phased array. The phased array is built on major challenge, as their electromagnetic performance is a flexible 4-layer polyimide and copper circuit board (total intimately tied to the physical dimensions and is thereby thickness ~12 mil). One side of the board holds the custom affected by shape modifications. For instance, the introduction CMOS IC, reference distribution network, and interface of a bend in a transmission line can adversely affect a circuit’s circuitry while radiators are mounted on the other side. The impedance matching or distort an ’s radiation pattern. radiator-side ground plane isolates the radiators from the other In the past, conformal design has been addressed by designing components. The components were positioned to maintain rigid, conformal RF systems for a specific, pre-determined flexibility around the axis along which the beam may be steered. shape [1]. Obviously, such rigid systems cannot conform to other shapes without a redesign and cannot be used in dynamically flexible systems such as electronics integrated with fabric. Phased arrays are particularly suited to flexible and conformal applications because they divide a given aperture into smaller elements, each experiencing small local deformation rather than a single large deformation [2]-[3]. These small deformations limit variations in the impedance matching at each element and the element patterns. However, the relative position of the elements must be ascertained to account for the shape deformation in such arrays, instead of calibrating radiator pattern and matching. This self-calibration is critical to proper operation of a flexible phased array system. A phased array can achieve the desired dynamic flexibility and conformability if it is electromagnetically insensitive to bending and capable of measuring and adjusting for changes in element position as the array deforms. For maximum utility, Fig. 1. Flexible Phased Array: (a) Circuit component side of flexible phased array. (b) In-plane view of flexible phased array. (c) Radiator side of flexible this shape calibration process should be self-contained, i.e., not phased array. reliant on external sensor(s) readout or other sources of information. In [4], an existing flexible phased array,

B. Bend Tolerant Radiator signal using an integrated power amplifier (PA) or convert a received signal to baseband through a direct down-conversion The ideal radiator for a conformal phased array should receiver. The functions of the IC are controlled through a digital possess a broad, near-hemispherical radiation pattern oriented interface. normal to the conformed surface and have low sensitivity to A large NMOS switch between the outer balun arm and bending in impedance matching and pattern. Ground plane ground allows the receiver to see a higher impedance when the backed dipoles satisfy these requirements for a 1-D array. Each PA is not operating. The receiver is unmatched but has radiator is made from a single polyimide and copper sheet sufficient dynamic range for the high-power inter-element soldered to the 4-layer phased array board, as shown in Fig. 2a. coupling measurements used for shape calibration. By keeping The element pitch is 18 mm (0.6λ at 10 GHz) when the array is the NMOS switch closed and enabling the PA, the RFIC is flat. The antenna feed connects to a single-ended coplanar capable of taking a self-loop measurement (measuring its own transmission line leading to the CMOS IC. The feed from the output), which is used in the calibration process. There are two board to the radiating arms acts as a balun to drive the dipole independent phase control systems in the IC, each separately differentially. offering greater than 2π phase control. The first is the PLL- The volume above the ground plane and directly below a based phase control that controls the phase of the transmitted dipole’s radiating arms does not experience significant signal and the receiver mixer LO. The second is an IQ phase deformation with bending. Because the strongest electric fields rotator before the PA, which only affects the transmitted phase. are contained in this region, the radiator is less sensitive to The phase rotator can be programmed from an on-chip SRAM, bending. Fig. 2b and Fig. 2c present the measured pattern of the allowing data to be rapidly modulated onto the transmitted 4th element in the array when the array is flat and when it is signal. The IC is implemented in TSMC’s 65nm CMOS process convexly conformed to a bend radius of 120 mm. The similarity and the die photo can be found in Fig. 3b. of the pattern with and without bending illustrates the radiators’ low sensitivity to bending.

Fig. 3. Transmit/Receive RFIC: (a) Integrated circuit block diagram. (b) Die photo.

III. COUPLING SHAPE CALIBRATION The proposed coupling shape calibration uses mutual coupling between elements to determine array shape. Mutual coupling has previously been used for phase calibration of rigid arrays [5] but has not been explored as a shape calibration technique. As a flexible phased array is deformed, the relative positions of its radiating elements changes. By measuring the distance between elements, the shape of the array can be reconstructed. Flexible phased arrays naturally adapt frequency Fig. 2. Radiator Element: (a) Radiators of the array are shown while the array modulated continuous wave (FMCW) radar as a distance is subjected to a 120mm bend radius deformation. (b) φ = 0 cut patterns from - measuring tool that can be implemented without significant 60° to +60° of the 4th element of the array, measured when the array is flat and when it is bent at a 120mm bend radius. (c) θ = 0 cut patterns. Both pairs of overhead to the array. patterns are normalized (dB) to the global maximum which occurs in the Bent Ideally, a bistatic radar measurement, which measures the Array, θ = 0 Cut. The simulated patterns are normalized to 0 degrees values of coupling path between two elements in the array, would be the flat measured results. The measured field is linearly polarized in the +θ sufficient to determine the distance between elements. However, direction. several non-idealities can hinder this measurement. In an array C. Transmitter/Receiver Integrated Circuit that is sufficiently dense to avoid grating lobes, the element The custom CMOS RFIC is responsible for generating and pitch is on the order of 0.5λ. This short distance limits the transmitting a high-power 10 GHz signal, as well as receiving amount of phase accumulated at baseband by the radar incident signals and down-converting them to baseband. The measurement between elements. Standard linear FMCW radar high-level block diagram of the IC is shown in Fig. 3a. The IC would observe a small fraction of a full period of the baseband uses a 2.5 GHz reference signal, which is converted to 10 GHz frequency corresponding to this return. Compounding the by an on-chip phase-locked loop (PLL) with a fixed 4x difficulty of measuring these short distances is the frequency- multiplication ratio. The output lock range of the PLL exceeds dependent phase response of active and passive components 9.9 GHz - 10.4 GHz. At 10 GHz the IC can either transmit the (such as amplifiers and transmission lines). The phase change

of these components from the beginning to the end of an RF Next, the calibration procedure measures the element A chirp is insignificant in typical FMCW applications where the self-loop to find 휃 (푓) as follows: baseband signal extends for many cycles, dwarfing their effect. 퐴퐴 However, in short distance measurements, the phase change of 휃퐴퐴(푓) = 퐻푡푥퐴(푓) − 퐻푟푥퐴(푓). (2) these components over the chirp bandwidth is comparable to After 휃 (푓) and 휃 (푓) are measured, the procedure the phase change of the radiation path. 퐵퐴 퐵퐵 Unlike standard FMCW measurements, the proposed calculates 퐻퐴퐵(푓) by calibration procedure measures total phase accumulated in the 퐻 (푓) = radiation path between the beginning and end frequencies of 퐴퐵 interest. This phase accumulation will be less than 2π for small ( 휃퐴퐵(푓) + 휃퐵퐴(푓) − 휃퐴퐴(푓) − 휃퐵퐵(푓))/2 (3) element separation, which is why phase is measured rather than After this set of measurements have been performed at 푓1 frequency. When this phase accumulation is less than a cycle, phase wrapping is not a concern, allowing unambiguous static and 푓2, 퐻퐴퐵(푓2) − 퐻퐴퐵(푓1) is calculated. single frequency measurements at multiple frequency points, To measure the shape of the flexible phased array, this for instance, at the beginning and end points of the chirp procedure can be performed between all elements in the array. bandwidth. This removes the need for a chirp-able phase If the distance between all elements is measured, the resultant reference signal and avoids challenges associated with chirp system of equations is likely over-defined. This additional non-linearity and PLL loop dynamics present in standard information can be used to reduce uncertainty. Furthermore, FMCW. For arrays large enough to experience phase wrapping some array deformations may break line-of-sight between two while measuring between elements, additional frequency points elements that are far apart in the array. For large-scale arrays along the bandwidth may be used. To account for components measuring all element pairs may be prohibitively time before and after the radiation path, the calibration procedure consuming. In these cases a subset of element pair requires self-loop measurements in which a single element acts measurements can be performed and used. This subset should as transmitter and receiver. This self-loop measurement can be be chosen to span the space of possible curvatures. subtracted from the coupling measurement to calibrate out the This calibration scheme assumes no significant objects in effect of these components from the measurement. the near-field of the array. It should be noted this calibration A simplified model of the self-calibration set-up is shown method does not account for changes in the phase response of the components before and after the transmit and receive path in Fig. 4. Transfer function, 퐻푡푥(푓), represents components only in the transmit path, such as the phase rotator and PA, split within each element such as the reference distribution network, antenna, and . These components can be while 퐻푟푥(푓) represents components only in the receive path, such as the LO buffer chain. The beginning and the end of the simulated or measured separately. Furthermore, the effect on the measurement of manufacturing variation in these measurement bandwidth is 푓1 and 푓2, respectively. The goal of components is mitigated by the measurements’ differential the calibration procedure is to measure the phase difference, nature: the calibration procedure measures the difference in

퐻퐴퐵(푓2) − 퐻퐴퐵(푓1), which is the phase accumulated over the phase at the beginning and end of the measurement bandwidth chirp bandwidth and is related to the distance between elements rather than absolute phase at any point. While demonstrated in a 1-D array, this calibration method can be naturally extended A and B. The procedure assumes 퐻퐴퐵(푓) is symmetric, as to 2-D and 3-D array calibration. traveled from A to B or B to A, knowing only passive elements lie in that path (reciprocity). IV. MEASUREMENTS A. Array Performance The beam-steering capability of the flexible phased array is demonstrated in Fig. 5. The figure shows the normalized gain pattern measurement for a steered beam when the array is flat and when it is conformed to a convex 120 mm bend radius. The element phases for beam-steering were determined through a coarse optimization procedure. The patterns are taken at the center of the bandwidth used for shape calibration Fig. 4. Simplified calibration model between elements A and B in the same flexible phased array. measurements, i.e. 10.176 GHz. B. Shape Calibration Measurement The calibration procedure first measures the phase of the coupling between the elements with A as the transmitter and B Shape calibration measurements were performed for the array when flat and when conformed to convex and concave as the receiver, 휃퐴퐵(푓), namely, bends. When bent, the phase centers of the radiators are moved farther apart or closer together. This change in distance is 휃퐴퐵(푓) = 퐻푡푥퐴(푓) + 퐻퐴퐵(푓) − 퐻푟푥퐵(푓). (1) measurable through shape calibration. The measurement is taken between 9.936 GHz and 10.416 GHz for a bandwidth of

radiation path for the element spacing listed on the x-axis. In addition to the measured data, three lines representing phase accumulation of a simplified free space model, where phase is linearly related with distance are plotted. The slope of the flat free space model is 10 degrees per element spacing as calculated using the bandwidth of the measurement and the element pitch. The y-intercept of the free space model is determined from an electromagnetic simulation of the coupling between adjacent elements in a flat array. The simulation found 49 degrees as the expected phase accumulation for adjacent Fig. 5. Beam-steering pattern measurements from -60° to +60° for flat and elements. The convex and concave bent measurements remain convexly conformed array. Each pattern is normalized to its highest measured value. above and below the flat measurements for all element spacings, respectively, correctly indicating the direction of the bend. 480 MHz. Each measurement is repeated for 24 phase rotator This measurement shows the viability of this approach for settings allowing for receiver nonlinearity correction [6]. flexible conformal array shape calibration. Adjacent element coupling was measured with the on-chip balun switch closed, but non-adjacent element coupling was V. CONCLUSION measured with the switch open to provide sufficient signal at This paper presents an 8-element, dynamically flexible and the receiver. This introduces a systematic difference between conformal, 1-D, phased array operating at 10 GHz. The array is the self-loop measurements and coupling measurements for assembled using a custom CMOS RFIC. The radiator elements non-adjacent elements. This is corrected by measuring the are insensitive to bending, and beam-steering capability is difference between the switch-open and switch-closed received demonstrated when the array is flat and when it is bent. A phase for each element while receiving from an adjacent method for flexible phased array shape calibration using the element, then subtracting this difference from the results. Fig. coupling between elements is proposed. The presented array is 6 presents the calibration measurement results for the array used to demonstrate the shape calibration method. when flat and when bent convexly and concavely to 120 mm bend radius. Because the deformation applied to the array is ACKNOWLEDGMENT symmetric, it is assumed the distance between each element The authors wish to thank their funders and collaborators at changes by a similar amount. With this assumption, the results the Caltech Space Solar Power Initiative, as well as M. R. for each possible element spacing (one element apart, two Hashemi, D. Hodge, C. Ives, and A. Ngai who aided in elements apart, etc.) are averaged together and presented as a measurement set-up and manuscript preparation. single data point. The y-axis of Fig. 6 shows the phase accumulated between 9.936 GHz and 10.416 GHz in the REFERENCES [1] Y. Zhao, Z. Shen, and W. Wu “Conformal SIW H-Plane on a Conducting Cylinder,” IEEE Antennas and Wireless Propagation Letters, vol.14, pp. 1271-1274, Feb. 2015. [2] P. Knott, T. Bertuch, H. Wilden, O. Peters, A. R. Brenner, and I. Walterscheid., “SAR Experiments Using a Conformal Radar Demonstrator,” International Journal of Antennas and Propagation, vol. 2012, Article ID 142542, 2012. [3] K.S. Beenamole, Sreejith C. A., G. Shankar, “Studies on conformal antenna arrays placed on cylindrical curved surfaces,” in 2016 International Symposium on Antennas and Propagation, 2016, p. 75. [4] B. Braaten, S. Roy, S. Nariyal, M. Aziz, N. Chamberlain, I. Irfanullah, M. T. Reich, and D. Anagnostou, “A Self-Adapting Flexible (SELFLEX) Antenna Array for Changing Conformal Surface Applications,” IEEE Transactions on Antennas and Propagation, vol. 61, pp. 655-665, Feb 2013. [5] H. M. Aumann, A. J. Fenn, and F. G. Willwerth, “Phased Array Antenna Calibration and Pattern Prediction Using Mutual Coupling Measurements,” IEEE Transactions on Antennas and Propagation, vol. 37, pp. 844-850, Jul 1989. [6] S.W. Ellingson. (2013) Correcting I-Q Imbalance in Direct Conversion Receivers webpage on Virginia Tech ECE. [Online]. Available: https://www.faculty.ece.vt.edu/swe/argus/iqbal.pdf

Fig. 6. Shape calibration measurements for bent and unbent array (120 mm bend radius). Linear regression best fit lines are plotted for each of the measured array shapes. Each measured data point represents the average of six separate measurements. The error bars show the standard deviation of the measurement for a given element separation. The convex measurements at element separations above 4 are not included because the bend has moved the elements out of line-of-sight.